JP7302203B2 - Passive sonar device, detection method, and program - Google Patents

Description

The present disclosure relates to a passive sonar device, its control method, and a program.

Generally, sonars are broadly classified into passive sonars and active sonars. A passive sonar is a device that receives sound waves emitted by a target in the sea and identifies the direction of the target from the direction of arrival of the sound waves. Active sonar, on the other hand, radiates sound waves into the sea, receives reflected sounds from targets, and calculates the target direction and distance from the time difference between the emitted and reflected sounds and the direction of arrival of the reflected sounds. is a device for identifying Note that the present disclosure is directed to acoustic signals acquired with passive sonar.

A general technique for identifying a sound source from an acoustic signal is to extract an acoustic feature from the result of frequency analysis of the acoustic signal and identify the sound source using a discriminator that receives the acoustic feature as an input. This technology is equally applicable to acoustic signals acquired under the sea by sonar. For example, Patent Documents 1 and 2 disclose the following techniques as techniques for identifying a sound source of an acoustic signal acquired under the sea using a discriminator.

The automatic target identification device disclosed in Patent Document 1 converts the frequency analysis result of the acoustic signal obtained in the sea into a code based on the characteristics peculiar to the sound source of the target, and uses a neural network with this code as an input to identify the target to identify the type of vessel.

The automatic target detection device disclosed in Patent Document 2 is a support vector machine that dimensionally compresses the frequency analysis result of an acoustic signal acquired in the sea or in the air for each time, converts it into a feature vector sequence, and uses this feature vector sequence as an input. to detect the presence of the target.

Japanese Patent No. 3159306 JP 2013-007578 A

However, the following two problems existed in the related technology for identifying the sound source of the acoustic signal acquired under the sea using a discriminator.
The first problem is that it is difficult to create a classifier with high accuracy. Since current sound source classifiers use statistical values of feature quantities, they require a large amount of data as acoustic signals for learning. Moreover, acoustic signals for training should not be selected randomly, but appropriate data should be selected so that the statistics represent the characteristics of the target sound source. If the acoustic signal for learning is inappropriate data, such as when the observation conditions are bad and ambient noise is mixed in, unnecessary components other than the components originally desired to be learned will be learned, resulting in a decrease in the accuracy of the classifier. It may lead to a decline.

A second problem is that it is difficult to respond to moving targets. When the target is moving, Doppler may occur depending on the position and speed of the target, and the frequency of the acoustic signal may change significantly (Doppler shift) over time. If a learning result that does not assume this case is used, detection omissions and detection errors of observed signals are more likely to occur. However, the presence or absence of Doppler shift differs depending on observation conditions, and the amount of change in frequency varies depending on the frequency of appearance. Therefore, it is generally difficult to deal with Doppler shift with a discriminator, which is a fixed learning result.

Accordingly, an object of the present disclosure is to provide a passive sonar device, a control method thereof, and a program that can solve any of the problems described above.

A passive sonar device according to one aspect comprises:
a receiver for acquiring underwater acoustic signals;
a logarithmic frequency analysis unit that performs frequency analysis of the acoustic signal on a logarithmic scale;
a basis selection unit for extracting basis vectors of a particular sound source from a database;
Using the frequency analysis result, which is the result of the frequency analysis, and the basis vectors, non-negative matrix factorization (NMF) is performed to calculate the activity of the basis vectors with respect to the acoustic signal. a signal separator;
a determination unit that detects that the acoustic signal is transmitted from the specific sound source using the activity of the basis vectors for the acoustic signal;
Prepare.

A method of controlling a passive sonar device, according to one aspect, comprises:
Acquiring underwater acoustic signals,
performing a frequency analysis of the acoustic signal on a logarithmic scale;
Extract the basis vectors of a particular sound source from the database,
Using the frequency analysis result, which is the result of the frequency analysis, and the basis vectors, non-negative matrix factorization (NMF) is performed to calculate the activity of the basis vectors with respect to the acoustic signal. ,
The activity of the basis vectors for the acoustic signal is used to detect that the acoustic signal was emitted from the particular sound source.

A program according to one aspect comprises:
to the computer,
a procedure for obtaining an acoustic signal in the sea;
performing a frequency analysis of the acoustic signal on a logarithmic scale;
a step of extracting basis vectors for a particular sound source from a database;
Using the frequency analysis result, which is the result of the frequency analysis, and the basis vectors, non-negative matrix factorization (NMF) is performed to calculate the activity of the basis vectors with respect to the acoustic signal. a procedure;
a step of detecting that the acoustic signal is emitted from the specific sound source using the activity of the basis vectors for the acoustic signal;
It is a program for executing

According to the above-described aspect, it is possible to provide a passive sonar device, a control method thereof, and a program that can contribute to solving the above-described problems.

1 is a diagram showing a configuration example of a passive sonar device according to an embodiment; FIG. It is a figure which shows the example of an operation|movement of the signal separation part which concerns on embodiment. It is a figure which shows the example of an operation|movement of the signal separation part which concerns on embodiment. FIG. 10 is a diagram showing a configuration example after a signal detection unit in a passive sonar device according to another embodiment; It is a figure which shows the structural example of the passive sonar apparatus which showed embodiment conceptually.

Embodiments of the present disclosure will be described below with reference to the drawings. Note that the following descriptions and drawings are appropriately omitted and simplified for clarity of explanation. Further, in each drawing below, the same elements are denoted by the same reference numerals, and redundant description is omitted as necessary.
<Embodiment>
FIG. 1 is a diagram showing a configuration example of a passive sonar device according to this embodiment.
As shown in FIG. 1, the passive sonar device according to the present embodiment includes a wave receiver 10 that acquires an underwater acoustic signal, and a class identification processing unit 20 that discriminates the sound source of the acoustic signal acquired by the wave receiver 10. , and a display unit 30 for displaying the identification result of the sound source identified by the class identification processing unit 20 . Further, the class identification processing section 20 includes a pre-processing section 21 , a signal detection section 22 and a post-processing section 23 . The pre-processing unit 21 also includes a logarithmic frequency analysis unit 211 , an orientation calculation unit 212 , a detection interval determination unit 213 , and a frequency analysis result division unit 214 . The signal detection unit 22 also includes a base database 221 , a base selection unit 222 , a signal separation unit 223 , a noise estimation unit 224 , and an activity normalization processing unit 225 . Further, the post-processing unit 23 includes a threshold determination unit 231 . In addition, in FIG. 1, the logarithmic frequency analysis unit 211 and the signal detection unit 22 indicated by thick lines are essential components of the present disclosure.

The operation of each component included in the class identification processing unit 20 will be described below.
[Pre-processing unit 21]
The logarithmic frequency analysis unit 211 receives the acoustic signal obtained by the wave receiver 10 as an input signal, performs frequency analysis on this input signal, and converts it into data expressed on a logarithmic frequency (octave) axis. . Frequency analysis on the logarithmic frequency axis is performed using Constant-Q Transformation (CQT). In this embodiment, on the premise of a configuration typically used in sonobuoys, acoustic signals observed by a directional sensor with directivity in the north-south direction and east-west direction, and an omnidirectional sensor with no directivity. A case where the acoustic signal observed by the sensor and the acoustic signal are input from the wave receiver 10 is shown. The frequency analysis result of the logarithmic frequency analysis unit 211 is output to the azimuth calculation unit 212 and the detection interval determination unit 213 .

The azimuth calculation unit 212 calculates the arrival azimuth of the input signal at each time and each frequency based on the frequency analysis result of the input signal. In the case where the wave receiver 10 forms an array arrangement with a plurality of sensors, the directionality synthesis is performed by a known method to obtain the incoming direction of the input signal. The processing result of the azimuth calculation unit 212 is output to the frequency analysis result division unit 214 .

Based on the frequency analysis result of the input signal, the detection section determination unit 213 estimates a section (no sound, mixture of excessive broadband noise) in which an abnormality has occurred in the communication path. This estimated section is handled in advance as being excluded from detection processing by the signal detection unit 22 . The processing result of the detection interval determination unit 213 is added to the frequency analysis result and output to the frequency analysis result dividing unit 214 .

Frequency analysis result dividing section 214 divides the frequency analysis result of the input signal into an arbitrary number based on the direction of arrival of the input signal. This processing emphasizes and suppresses arbitrary directions, and obtains a frequency analysis result with an improved S/N ratio (Signal-Noise ratio). Specifically, the frequency analysis result division unit 214 sets azimuth sectors obtained by equally dividing the azimuth of 360 degrees by the number of divisions, and repeats the process of extracting data including the arrival azimuth in the azimuth sectors. For example, if the number of divisions is 8, the first data is data in which the components other than those in the direction of 0±α degrees are replaced with 0, and the second data is data in which the components other than those in the direction of 45±α degrees are replaced with 0. This process is repeated to obtain eight frequency analysis results divided for each azimuth. α is a parameter and is usually set so that adjacent azimuth sectors overlap. The frequency analysis result for each azimuth obtained by the frequency analysis result division unit 214 is output to the basis selection unit 222 of the signal detection unit 22 .

[Signal detector 22]
The signal detection unit 22 separates the input signal based on the input frequency analysis result for each azimuth. In this embodiment, a case where non-negative matrix factorization (NMF: Non-negative Matrix Factorization) is used as a method of signal separation processing will be described. NMF is a technique for searching for arbitrary vectors (called bases) from data (images, sound signals, etc.) represented by matrices as a product of a plurality of vectors. As an output of the NMF, a vector (referred to as "activity") is obtained which indicates "at what time" and "with what intensity" the vector to be searched exists in the data.

It is assumed that one or more pieces of frequency characteristic information of sound sources to be searched are registered in advance in the base database 221 as base vectors for one sound source.

The basis selection unit 222 extracts the basis vectors of the search target sound source from the basis database 221 . When two or more basis vectors corresponding to one sound source are registered in the basis database 221, the basis selection unit 222 selects basis vectors that satisfy the following two conditions. The first condition is that the product of the basis vector and the activity vector approximates the frequency analysis result, that is, the distance between the basis vector and the activity vector is sufficiently small. The second condition is that the basis vectors have a simple (sparse) structure represented by as few frequencies as possible. If the basis vectors have a sparse structure, that is, basis vectors that do not have unnecessary components, the activation level is not affected even if some frequencies of the sound source do not appear, and detection omissions can be prevented. . We express these two conditions as a function and estimate the basis vectors by optimizing the function. The basis vectors extracted by the basis selection section 222 are output to the signal separation section 223 together with the frequency analysis result for each direction.

The signal separation unit 223 performs NMF using the frequency analysis result for each azimuth and the basis vector, and calculates the activity of the basis vector of the sound source to be searched for the input signal.
Here, processing of the signal separation unit 223 will be described in detail with reference to FIGS. 2 and 3. FIG. 2 and 3 are examples in which the search range of the basis vector is ±3 bins of the basis vector.

FIG. 2A represents the basis vectors of the sound sources to be searched. Also, in FIG. 2A, the basis vectors exist at the fundamental frequency of the sound source to be searched and at the frequencies of overtones of the fundamental frequency. Also, in FIG. 2A, the frequency axis of the basis vectors is expressed in logarithmic scale.

(B) of FIG. 2 represents the frequency analysis result of a certain direction. Also, in FIG. 2B, the frequency axis of the frequency analysis result is represented by a logarithmic scale. In FIG. 2B, cells with high density represent acoustic signals of the sound source to be searched, and cells with low density represent acoustic signals of other sound sources around the sound source to be searched. .

As shown in FIG. 2B, the acoustic signal of the sound source to be searched has its frequency shifted to the low frequency side with time (that is, Doppler shift). It can be seen that acoustic signals are generated in -2 bin to +2 bin).

(C) of FIG. 2 represents an activity matrix obtained from the basis vectors of (A) of FIG. 2 and the result of frequency analysis of (B) of FIG. Each column of this activity matrix is a column vector obtained by calculating the inner product between the basis vector and each column of the frequency analysis result. This inner product calculation corresponds to NMF.

More specifically, for the purpose of following the Doppler shift, the activity matrix shifts the basis vectors in an arbitrary range (here, ±3 bins of the basis vectors) in the frequency direction, and the basis vectors and the frequency analysis result This is the result of calculating the inner product of . The reason why the basis vectors can be shifted in the frequency direction is that the frequency axis is represented by a logarithmic scale.

In the activity matrix of FIG. 2C, the inner product calculated without shifting the frequency of the basis vectors is a column vector of "0", and the inner product calculated by shifting the frequency of the basis vectors by +2 bins is a column of "+2". vector. Here, since the search range of the basis vectors is ±3 bins of the basis vectors, the activity matrix is composed of a total of 7 column vectors from −3 bins to +3 bins.

Thus, in the present embodiment, signal separation section 223 shifts the basis vectors in the frequency direction within an arbitrary range, calculates the inner product of the basis vectors and the frequency analysis result, and calculates the activity matrix.

If the basis vectors are not shifted in the frequency direction, the activity matrix contains only 0-bin information. Therefore, as shown in FIG. 2B, when the acoustic signal of the sound source to be searched is generated in a wide frequency band (-2 bin to +2 bin) due to the influence of Doppler shift, -2, -1 , +1, +2 bin information is lost.

On the other hand, when the basis vectors are shifted in the frequency direction within an arbitrary range as in this embodiment, the acoustic signal of the sound source to be searched is spread over a wide frequency band (-2 bins to +2 bins) due to the influence of Doppler shift. Even if this occurs, it is possible to suppress loss of information (information of −2, −1, +1, +2 bins).

As described above, signal separation section 223 calculates an activity matrix for each direction. The activity matrix for each direction is output to noise estimation section 224 .

Next, as shown in FIG. 3, the signal separation unit 223 extracts the maximum value at each time from the activity matrix as a component that may be the activity for the input signal, and converts the extracted value to Calculate activity vectors arranged in the time direction.

As described above, the signal separation unit 223 calculates an activity vector for each direction. The activity vector for each orientation is output to the activity normalization processing section 225 .

The noise estimation unit 224 estimates the ambient noise for each direction so that the activity normalization processing unit 225 normalizes the activity vector for each direction. The noise estimating unit 224 extracts activities other than the maximum value at each time as activities against ambient noise from the input activity matrix, and calculates a noise activity vector in which the extracted activity values are arranged in the time direction. . Thus, the noise estimator 224 calculates the noise activity vector for each direction. The noise activity vector for each direction is output to activity normalization processing section 225 .

The activity normalization processing unit 225 normalizes the activity vector for each direction using the noise activity vector. Next, activity normalization processing section 225 concatenates the normalized activity vectors for each orientation to form a normalized activity matrix, and from this normalized activity matrix, the maximum value at each time is calculated as The activity is extracted, and a one-dimensional normalized activity vector is calculated by arranging the extracted activity values in the time direction. This normalized activity vector is output to the threshold determination section 231 .

[Post-processing unit 23]
The threshold determination unit 231 compares the numerical value of the normalized activity vector at each time with the detection threshold. If the numerical value of the normalized activity vector at each time is equal to or greater than the detection threshold, the threshold determination unit 231 determines that the sound source to be searched corresponding to the basis vector exists at that time, that is, the input signal is emitted from the sound source to be searched.

As described above, according to the present embodiment, the acoustic signal in the sea acquired by the wave receiver 10 is frequency-analyzed on a logarithmic scale, and the NMF is performed to compute the activity of the basis vectors for the acoustic signal, and the activity is used to detect that the acoustic signal was emitted by the sound source being searched.

Here, algorithms based on machine learning represented by deep learning express knowledge information corresponding to the basis in the present embodiment by combinations of various factors, not limited to frequencies. This is a measure to improve the accuracy of highly difficult classification and recognition, but in order to obtain such knowledge information, it is necessary to collect a large amount of data through human hands with experience and expertise. and appropriate data extraction work is required.

On the other hand, since NMF uses only the basis, that is, frequency as knowledge information, the amount of data for learning is much smaller than deep learning. Also, if the frequency of the sound source to be searched is known, it is possible to use that frequency as the basis.

Therefore, by using NMF for the signal separation processing method as in the present embodiment, a large amount of data is not required as the acoustic signal for learning, and the acoustic signal for learning is inappropriate data. As a result, it is possible to prevent the accuracy of the class identification processing unit 20 from deteriorating.

Further, according to the present embodiment, the basis vectors are shifted in an arbitrary range in the frequency direction, the activity matrix is calculated as a result of calculating the inner product of the basis vectors and the result of frequency analysis, and each Extract the maximum value of time as activity.

As in this embodiment, by shifting the basis vectors in an arbitrary range in the frequency direction, even if the acoustic signal of the sound source to be searched is generated in a wide frequency band due to the influence of Doppler shift, loss of information can be suppressed. Therefore, even if a Doppler shift occurs in the acoustic signal, the class identification processing unit 20 can cope with the Doppler shift.

<Other embodiments>
As a data division method in the frequency analysis result dividing unit 214, a method of emphasizing a component in a specific direction by beam forming and suppressing a component in a true opposite direction may be used. In this case, the frequency analysis results after division have components in all directions (even components in the exact opposite directions are suppressed, but they do not become 0). The difference between degrees increases. This is advantageous in the threshold decision section 231 that decides that a sound source has been detected. Furthermore, by combining known directivity synthesizing means with the calculation of the direction of arrival by the direction calculation unit 212, the S/N ratio may be optimized.

In addition, under the premise that the association between targets (ships, submarines, etc.) and sound sources is stored in a database, multiple sound sources are searched as follows, that is, multiple basis vectors corresponding to multiple sound sources are generated. You may carry out a search. For example, when an arbitrary sound source #1 is detected, the target having the sound source #1 is searched from the database, and the base vector of the other sound source #2 of the searched target is extracted from the base database 221 . Then, with the basis vector of this sound source #2 as an input, the processing after the signal separation unit 223 of the signal detection unit 22 is executed.

Also, after multiple sound sources are detected, as shown in FIG. 4, target identification may be performed using a combination of multiple sound source detection results.
In the example of FIG. 4, a plurality of signal detectors 22-1 to 22-n are provided corresponding to a plurality of sound sources #1 to #n (where n is an integer of 2 or more). -1 to 22-n, a target determination unit 24 is provided in the subsequent stage.

The signal detection units 22-1 to 22-n have functions similar to those of the signal detection unit 22. FIG.
For example, the signal detection unit 22-1 extracts the basis vector of the sound source #1 corresponding to itself, and then receives the extracted basis vector of the sound source #1 as an input, and executes the processes after the signal separation unit 223. As a result, the signal detector 22-1 outputs a normalized activity vector for the basis vectors of the sound source #1 with respect to the input signal.
Similarly, the signal detector 22-n outputs a normalized activity vector for the basis vector of the sound source #n with respect to the input signal.

The target determination unit 24 can detect the existence of the sound sources #1 to #n, for example, by the same method as the threshold value determination unit 231. FIG. The target determination unit 24 also holds a discriminator that has learned the relationship between the sound source and the target (for example, A is the target having the sound source #1 and the sound source #2). Therefore, the target determination unit 24 uses this discriminator to discriminate the target based on the detection result of each sound source. The identification result of the target determination unit 24 is output to the display unit 30 and displayed. Note that the target determination unit 24 may be provided instead of the threshold value determination unit 231 or may be provided in addition to the threshold value determination unit 231 .

<Concept of Embodiment>
Next, the concept of the passive sonar device according to the embodiment described above will be described.
FIG. 5 is a diagram showing a configuration example of a passive sonar device conceptually showing the embodiment described above.

The passive sonar apparatus shown in FIG. 5 includes a wave receiver 91 , a logarithmic frequency analysis section 92 , a base selection section 93 , a signal separation section 94 and a determination section 95 .
The wave receiver 91 acquires an underwater acoustic signal. A wave receiver 91 corresponds to the wave receiver 10 .

A logarithmic frequency analysis unit 92 performs frequency analysis of the acoustic signal on a logarithmic scale. The logarithmic frequency analysis section 92 corresponds to the logarithmic frequency analysis section 211 .
A basis selection unit 93 extracts basis vectors of a specific sound source from the database. The basis selection section 93 corresponds to the basis selection section 222 .

The signal separator 94 uses the frequency analysis result and the basis vectors to perform non-negative matrix factorization (NMF) to calculate the activity of the basis vectors for the acoustic signal. The signal separator 94 corresponds to the signal separator 223 .

The determination unit 95 detects that an acoustic signal is transmitted from a specific sound source using the activity level of the basis vectors for the acoustic signal. A determination unit 95 corresponds to the threshold determination unit 231 or the target determination unit 24 .

Note that the signal separation unit 94 shifts the basis vectors in an arbitrary range in the frequency direction, calculates an activity matrix that is the result of calculating the inner product of the basis vectors and the frequency analysis result, and from the calculated activity matrix The maximum value at each time may be extracted as the activity, and an activity vector may be calculated by arranging the extracted maximum values at each time in the time direction.

Also, the passive sonar device may further include a frequency analysis result dividing unit (corresponding to the frequency analysis result dividing unit 214) that divides the frequency analysis result for each azimuth. In this case, the signal separation unit 94 may calculate an activity matrix for each direction and an activity vector for each direction. In this case, the activity vector for each orientation is connected to calculate a normalized activity matrix, the maximum value at each time is extracted from the normalized activity matrix, and the extracted maximum value at each time is calculated as may further include an activity normalization processing unit (corresponding to the activity normalization processing unit 225) that calculates the normalized activity vectors arranged in . Also, in this case, the determination unit 95 may compare the normalized activity vector with a threshold to detect that the acoustic signal is transmitted from a specific sound source.

Also, there may be a plurality of specific sound sources. In this case, the basis selection section 93 and the signal separation section 94 may be provided for each specific sound source. Further, in this case, the basis selection unit 93 extracts the basis vectors of the specific sound source corresponding to itself from the database, and the signal separation unit 94 uses the basis vectors of the specific sound source corresponding to itself and the frequency analysis result. Then, the NMF is executed to calculate the activity of the basis vectors for the acoustic signals, and the determination unit 95 may detect that the acoustic signals are transmitted from a plurality of specific sound sources.

In addition, when detecting that the acoustic signal is transmitted from a plurality of specific sound sources, the determination unit 95 uses a discriminator that has learned the relationship between the sound source and the target to detect the plurality of specific sound sources. A target may be identified.

Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

For example, in the above embodiments, the passive sonar device of the present disclosure has been described as a hardware configuration, but the present disclosure is not limited to this. According to the present disclosure, arbitrary processing of the passive sonar device can be realized by a processor such as a CPU (Central Processing Unit) reading and executing a computer program stored in memory.

In the above examples, the programs can be stored and delivered to computers using various types of non-transitory computer readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (eg, flexible discs, magnetic tapes, hard disk drives), magneto-optical recording media (eg, magneto-optical discs), CD-ROMs (Compact Disc-Read Only Memory), CDs - R (CD-Recordable), CD-R/W (CD-ReWritable), semiconductor memory (e.g. mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) include. The program may also be delivered to the computer by various types of transitory computer readable medium. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. Transitory computer-readable media can deliver the program to the computer via wired channels, such as wires and optical fibers, or wireless channels.

REFERENCE SIGNS LIST 10 wave receiver 20 class identification processing unit 21 pre-processing unit 211 logarithmic frequency analysis unit 212 direction calculation unit 213 detection interval determination unit 214 frequency analysis result division unit 22 signal detection unit 221 base database 222 base selection unit 223 signal separation unit 224 noise Estimation unit 225 Activity normalization processing unit 23 Post-processing unit 231 Threshold determination unit 24 Target determination unit 30 Display unit 91 Wave receiver 92 Logarithmic frequency analysis unit 93 Base selection unit 94 Signal separation unit 95 Determination unit

Claims (6)

a receiver for acquiring underwater acoustic signals;
a logarithmic frequency analysis unit that performs frequency analysis of the acoustic signal on a logarithmic scale;
a basis selection unit for extracting basis vectors of a particular sound source from a database;
An activity matrix is calculated by shifting the basis vectors in an arbitrary range in the frequency direction, calculating an inner product of the basis vectors and the frequency analysis result, which is the result of the frequency analysis. a signal separation unit that extracts the maximum value of time and calculates an activity vector in which the extracted maximum value of each time is arranged in the time direction;
Judgment for comparing the numerical value of the activity vector at each time with a threshold value, and detecting that the acoustic signal is transmitted from the specific sound source if the numerical value at each time of the activity vector is equal to or greater than the threshold value. Department and
A passive sonar device comprising: further comprising a frequency analysis result dividing unit that divides the frequency analysis result by direction,
The signal separation unit calculates the activity matrix for each orientation and calculates the activity vector for each orientation,
A normalized activity matrix is calculated by connecting the activity vectors for each orientation, the maximum value at each time is extracted from the normalized activity matrix, and the extracted maximum values at each time are arranged in the time direction. further comprising an activity normalization processing unit that calculates a normalized activity vector,
The determination unit compares the numerical value of the normalized activity vector at each time with a threshold value, and if the numerical value of the normalized activity vector at each time is equal to or greater than the threshold value, the acoustic signal corresponds to the specific sound source. to detect that it has been sent from
A passive sonar device according to claim 1 . There are a plurality of said specific sound sources,
The base selection unit and the signal separation unit are provided corresponding to the specific sound sources, respectively;
The basis selection unit extracts the basis vectors of the specific sound source corresponding to itself from the database,
The signal separation unit calculates the activity vector using the basis vector and the frequency analysis result of the specific sound source corresponding to itself,
The determination unit detects that the acoustic signals have been transmitted from a plurality of the specific sound sources,
3. A passive sonar device according to claim 1 or 2 . When detecting that the acoustic signals are transmitted from the plurality of specific sound sources, the determination unit identifies the plurality of detected specific sound sources using a discriminator that has learned the relationship between the sound source and the target. identify goals that have
4. A passive sonar device as claimed in claim 3 . A detection method using a passive sonar device,
Acquiring underwater acoustic signals,
performing a frequency analysis of the acoustic signal on a logarithmic scale;
Extract the basis vectors of a particular sound source from the database,
An activity matrix is calculated by shifting the basis vectors in an arbitrary range in the frequency direction, calculating an inner product of the basis vectors and the frequency analysis result, which is the result of the frequency analysis. Extract the maximum value of the time, calculate the activity vector by arranging the maximum value of each extracted time in the time direction ,
Comparing the numerical value of each time of the activity vector with a threshold value, and detecting that the acoustic signal is transmitted from the specific sound source if the numerical value of each time of the activity vector is equal to or greater than the threshold value;
Detection method. to the computer,
a procedure for obtaining an acoustic signal in the sea;
performing a frequency analysis of the acoustic signal on a logarithmic scale;
a step of extracting basis vectors for a particular sound source from a database;
An activity matrix is calculated by shifting the basis vectors in an arbitrary range in the frequency direction, calculating an inner product of the basis vectors and the frequency analysis result, which is the result of the frequency analysis. A procedure of extracting the maximum value of time and calculating an activity vector in which the extracted maximum value of each time is arranged in the time direction;
A step of comparing the numerical value of the activity vector at each time with a threshold, and detecting that the acoustic signal is transmitted from the specific sound source if the numerical value of the activity vector at each time is equal to or greater than the threshold. and,
program to run the

Images